- 1 -

2006 Florida Conference on Recent Advances in Robotics, FCRAR 2006 Miami, Florida, May 25-26, 2006

Fuzzy Logic Based Integrated Controller for Unmanned
Aerial Vehicles
Kimberly Bickraj,
Thierry Pamphile
Florida International University
Department of Mechanical and
Materials Engineering
10555 West Flagler Street,
Miami, FL 33174
(305) 348-3770
kbick001@fiu.edu
tpamp001@fiu.edu


Aylin Yenilmez
Istanbul Technical University
Faculty of Mechanical
Engineering
Gumussuyu, Istanbul 34439
Turkey
90 (212) 293-1300/2448
yenilmezay@itu.edu.tr
Ming Li, Ibrahim N. Tansel
Florida International University
Department of Mechanical and
Materials Engineering
10555 West Flagler Street,
Miami, FL 33174
(305) 348-3304
mli002@fiu.edu
tanseli@fiu.edu

ABSTRACT
This paper discusses the integration of health monitoring and
flight control systems for small Unmanned Aerial Vehicles
(UAVs). After briefly reviewing previous fuzzy logic controllers
(FLC) of air vehicles, a very low cost integration method is
proposed. The proposed fuzzy logic (FL) selects the best gain
values for the operation of PD or PID controllers of the
autonomous flight system according to the health of the
components. Such gain adjustments help the UAV to execute
maneuvers in a more conservative manner when the system have
structural or proportion system problems.
Keywords
Unmanned Aerial Vehicles, Fuzzy Logic Control, ANFIS,
Neural Network
1. INTRODUCTION
The latest flight missions have demonstrated the benefits of
small UAVs. They are cheaper, do not risk the lives of the
pilots, are more difficult to inflict damage on to than
conventional planes because of their size, and can be operated
with much smaller fighter teams than what is typically required.
As technology progresses, the capabilities and the reliability of
the UAVs will improve and they will be used for more
demanding tasks. New UAV designs have already been
discussed for combat and space applications. In order to assess
the condition of the vehicle and to operate the system safely in
the event of any component failure, more complex health
monitoring systems will be required.
Today, the UAVs employed are operated from remote locations
or by programmed waypoints. The ultimate goal is to develop
fully independent UAVs, which can conduct specialized and/or
covert operations for warfare, weather and disaster surveillance,
imaging spectrometry, and reconnaissance [1]. With the
implementation of such a control system, the small UAV will
not be limited to visual range or short distances for operation nor
will it be dependent on human control from an operational base.
At present, the forecasted optimal method for such small UAV
focuses on a hybrid scheme, which involves the use of neural
networks and fuzzy logic control [2]. The neural network could
be used to learn the pilots skill in maneuvering a flight vehicle.
FL is very convenient to develop sophisticated hardware that
directly implements engineers knowledge. FL will be used to
tune the PD or PID control gains according to the readings of the
sensors, which monitors the condition of the structure and the
engine.
Since the development of such a hybrid system proves to be
time consuming and involves generating many algorithms and
models, we focus only on one strand of the hybrid scheme,
which is the fuzzy logic control in this paper. This artificial
intelligence system is quite favored for automatic control
because it avoids complex non-linear equations and can utilize
the best pilot expertise available. Fuzzy logic, which is based on
the mathematical theory of fuzzy sets [3], circumvents complex
differential equations by offering a collection of if-then rules
which operates as a linear function even though the function
itself is not known. Moreover, this logic does not operate on
binary output such as true or false, up or down, left or right, but
rather facilitates for the entire intermediate spectrum of outputs
to be included. In this paper, previous fuzzy logic controllers
(FLC) are reviewed and a simple integration method is
proposed. This method correlates the gain and the limits of the
actuators for conservative vehicle operation when some of the
components have problems.
2. FUZZY LOGIC CONTROL
2.1. Background
Prior to investigating the fuzzy logic cases, we examine the
Fuzzy Logic Control (FLC) implementation process and specific
modes of implementation. There are two types of fuzzy logic
inference systems (FIS) available: Mamdani and Sugeno. Both
of these methods essentially contain the same steps, the main
difference being that the Sugeno method outputs are usually
linear. The following FLC cases mainly utilize the Mamdani
method. In terms of the fuzzy logic processes, there are three
main processes, shown in Figure 1. In the first main
fuzzification process, a crisp value is fuzzified which means
that it is characterized by a fuzzy set. This is followed by the
inference process whereby the fuzzified inputs that are
represented in the form of membership functions are combined
and used to display the knowledge of the experts in the form of
rules. In the final process known as defuzzification, the output
- 2 -
2006 Florida Conference on Recent Advances in Robotics, FCRAR 2006 Miami, Florida, May 25-26, 2006

value is defuzzified thus allowing obtaining the important
crisp value. In the cases studied, the center of area
defuzzification method is employed which generates the center
of gravity of the output set. The membership functions
mentioned earlier are vaguely defined sets, which represent the
range of the various inputs. There geometries can be triangular,
trapezoidal, singleton or variations of the Gaussian function, etc.
Genetic algorithms and neural networks [4] may be used to
develop and tune the functions.

Figure 1. Main processes of fuzzy logic method
2.2. Previous Applications
With the fuzzy logic methodology kept in mind, we examined a
few cases of fuzzy logic implementation in flight control. The
first fuzzy system analyzed was taken from Mengalis paper.
This simulation involved a two loop system (see Figure 2) which
made use of a reference model plane (an ideal model whereby
the airplane dynamics have been linearized about a point in the
flight envelope and driven by a reference input r) versus a real
plane whose dynamics are nonlinear. Thus in this system, the
inner loop involved the application of the reference signal R to
both systems. This generated an error as the performance of the
realistic model did not match the optimized performance of the
ideal model (due to model uncertainties, nonlinearities etc) [4].
Applying this error signal to the fuzzy knowledge based
controller, the supervisory expert control system (SECS) was
employed. This proportional derivative controllers
responsibility was to drive the error of the real model to zero. Its
output U was added to the real model to obtain the performance
of the outer loop.

Figure 2. Loop arrangement of the aircraft control system
In the second case study, the controller of a 1m small-size UAV
was developed using fuzzy logic and its integration was then
tested [1]. Two control methods were used. The first approach
collected information about the inputs of the pilot using a
learning algorithm. The plane used a PD controller and two
fuzzy logic combinations to allow the plane to maintain its
altitude. The PD selected the most appropriate control surface
displacements. The two separate fuzzy logics adjusted the
proportional and derivative gains by considering the pitch and
change angle errors. Seven levels were used with the fuzzy
logic. The flight tests demonstrated that the fuzzy logic closed-
loop control worked better than the opened-loop control.
According to Rajasekhar and Sreenathas [5], fuzzy logic can be
used to study missile guidance control. The FLC system was
developed using the PNG law
c
=NV
c
where
c
is the
achievable control, V
c
is the line of sight rate (LOS), is the
closing velocity and N is a navigation constant. Hence, the goal
of the controller was to zero the LOS rate by making the missile
turn rate be directly proportional to the LOS rate [2].

Figure 3
represents how the variables were utilized in this FLC system.

Figure 3. Missile Guidance arrangement using a fuzzy logic
controller
Finally, we examined the application of unmanned landing of an
aircraft [3]. This study focused on the nonlinear control actions
of a fuzzy controller (FC) operating effectively in the automated
landing of unmanned aircrafts. A fuzzy inference system was set
up with two inputs and one output. Seven triangular membership
functions were used for the inputs to be converted into
linguistics variables. Seven triangular membership functions
were also used for the output functions and forty-nine fuzzy
logic rules were established. The fuzzy controller successfully
controlled three-dimensional flight path, sink rate, and angular
altitudes to allow safe landing of the unmanned aircraft. A High
SIQ (System Intelligent Quotient) mechanism was created
which combined elements of human pilot intelligence with
landing knowledge. The system was tested with simulations as
well as in "real-life" landings in which operations demonstrated
definite potential of fuzzy logic usage as a controller.
3. FUZZY LOGIC IMPLEMENTATION
After studying the fuzzy logic implementations for aircraft
modeling, projectile guidance, altitude maintenance and
unmanned landing, we now put forth our strategy for a flight
control system. For a small UAV, basic navigation (in horizontal
and vertical planes) utilizing FLC will be attempted. In terms of
optimizing parameters, as we mentioned before neural networks
are good at learning and recognizing patterns however, their
training and learning processes require a lot of time.
Furthermore, it is difficult to analyze the trained network and
explain how neural networks get their decisions. In order to
acquire our control rules and tuned membership functions we
relied on Shiori and Uenos approach [6]. Input and output
elements are selected based on PD control. The two input
variables utilized are the offset distance and the difference of the
directional angle, both whose membership functions are
triangular. Thus, based on the difference of the coordinates and
the directional angle, the expert controllers responsibility will
be to drive these errors to zero accordingly by altering the
throttle as well as the aileron, rudder and elevator positions.
Additionally the control output will also employ a triangular
membership function. As rule bases are given for course
Normalization Fuzzification
using
Membership
Fxn
Fuzzy Inference
System using
rules
Defuzzification
using center-of-
gravity
Denormalization

V
c


FLC system

Missile

c

Ideal
model
Real
model
SECS
e

-
+
R
U
- 3 -
2006 Florida Conference on Recent Advances in Robotics, FCRAR 2006 Miami, Florida, May 25-26, 2006

maintenance in both the horizontal and vertical direction from
Shiori and Uenos paper, we generate two different fuzzy logic
systems shown in Tables 1 and 2, each of them containing their
own separate rules.
Table 1. Horizontal direction
MF of Closing offset dist
(hor)
MF of Error angular diff

MF of control output 3D Surface


View of Rules


In terms of flight guidance, the control systems of UAVs usually
follow the instructions of the pilots at remote location. The more
sophisticated systems follow the programmed waypoints with
the help of their GPS to reach to their targets. However, in order
to maximize the performance of the UAVs tasks, adjustment of
the sophisticated systems flight characteristics according to the
operating and structural condition would be quite beneficial.
Hence, an ideal Integrated Structural Health Monitoring and
Autonomous Guidance and Control (ISHM + AG&C)
integration is presented in the Figure 4. It evaluates the flight
path, flight condition, and structural health information
simultaneously. It processes the gathered information and
determines the proper flight characteristics. It may report the
abnormalities to the ground crew. The architecture of a simple
integrated system is presented in Figure 5.
Table 2. Vertical direction
MF of Closing offset dist
(vert)
MF of Error angular diff

MF of control output 3D Surface


View of Rules

- 4 -
2006 Florida Conference on Recent Advances in Robotics, FCRAR 2006 Miami, Florida, May 25-26, 2006


Figure 4. The integrated ISHM and AG&C

Figure 5. The simplest approach for the integrated ISHM
and AG&C system
A fuzzy, fuzzy-neural, or a neural network based FC and SHM
Interface to AG&C unit (Figure 6) can be used to evaluate
flight condition and structural health data. Flight condition data
is primarily the engine performance information including the
vibratory signature, emission, and temperature of the engine(s).
The structural health data is the location, size and severity of
structural problems. The signal-processing unit may determine
the gain to control the severity of the control surface actions.
Ideally, the gain should have a nonlinear relationship with the
inputs. Fuzzy logic is a fine choice if the developers want to
provide rules. If the developers want to implement an accurate
and sophisticated IO relationship, the membership functions and
rules can be established using genetic algorithm.
The information coming from the ISHM is processed using
fuzzy logic. The gains and the ranges of the AG&C are
controlled based on the condition of the structure, control
surfaces and engine. A simple implementation of the FC and
SHM interface is presented in the Figure 6. The corresponding
FL system is presented in the Table 3.

Table 3. A simple fuzzy logic controller to adjust the agility
of UAVs according to flight and structural conditions








FC and
ISHM
Interface
to AG&C
Determine
control
surface
positions
Flight
condition
Structural
health
monitoring
Position
related
information
Reports
to the
base
Signals to
actuators
Gains
or
range



Integrated SHM +
AG&C

Actuators of
the control
surfaces
Flight
condition
Structural
health
monitoring
Position related
information
Reports to the
base
- 5 -
2006 Florida Conference on Recent Advances in Robotics, FCRAR 2006 Miami, Florida, May 25-26, 2006

Table 3. A simple fuzzy logic controller to adjust the agility
of UAVs according to flight and structural conditions
(continued)




Figure 6. An integrated flight control and integrated system
health monitoring system concept

4. RESULTS AND DISCUSSIONS
The input and output relationship of the FL were presented with
3-D plots in the Tables 1and 2. The smoothness, consistency and
representation of desired characteristics strongly suggest the
feasibility of the FL applications for control of UAVs and
integration of FC and SHM. Program size was very reasonable.
Pentium microprocessor of our microcomputers calculated the
input output relationship and created the corresponding plots in
a fraction of second for a 25X25 grid. The computations were
completed in two to three seconds when the grid size was
increased to 200X200. FL has been implemented using
microprocessors and digital signal processors (DSP) in many
engineering applications. All these observations indicated that
FL could be effectively used for integration of FC and SHM
systems. The integrated system could control the agility of the
plane by changing the gains of the PD and PID controllers.
Another option for implementation of the desired characteristics
to the integrated system is to use the adaptive neuro-fuzzy
inference system (ANFIS). Performance of ANFIS was
evaluated by following Yingjies approach in his paper on Auto-
landing [7]. In his study, two inputs including the altitude and
velocity and one output (force) were utilized. As the ANFIS
system only supports Sugeno type fuzzy system, the Mamdani
data first had to be imported and an initial FIS had to be
generated. ANFIS then attempts to learn from the data set to get
the membership function and rules according to the
characteristics of the initial FIS (see Table 4). The architecture
of the ANFIS is presented in Table 4. There are three layers with
multiple nodes between the two inputs and one output.
Another important difference of the ANFIS system from the
conventional fuzzy logic is having only one output function that
has to be linear or constant. There is no sharing of rules for the
output membership functions [8]. The membership function of
the output of our case study is presented in the Table 5.
However, after ANFIS completed all the training and the
generation of rules, the 3-D plots, which show the relationship
between the inputs and the output, was very similar to the
desired characteristics in Table 6. In the event that the network
was unable to represent all the features of the system then
another set of testing data would be required to help model
validation. As a result, an overall important note about using the
neural network is that initial data still has to be given to the
system and it must be capable of developing an acceptable
algorithm within a reasonable time. Once this is successfully
established, we will be one-step closer to achieving a completely
self-sufficient UAV.
Table 4. Fuzzy Interface System (FIS) and nodal network
Training Data Nodal Network


Table 5. Membership Functions
Membership Function of
Force FIS
Membership Function of Force
for ANFIS




FC and ISHM
Interface to
AG&C
Rudder gain
Aileron gain
Elevator gain
Engine throttle
Wing condition
Engine condition
Body condition
Rudder condition
Aileron condition
Elevator
condition
Rules and typical surface
- 6 -
2006 Florida Conference on Recent Advances in Robotics, FCRAR 2006 Miami, Florida, May 25-26, 2006

Table 6. Control Surfaces
Control Surface for FIS Control Surface for ANFIS
5. CONCLUSION
In this paper, we reviewed some of the well-known studies on
the use of fuzzy logic (FL) for the flight control path of UAVs.
The study indicated that FL is very suitable for the task and can
be easily implemented using even simple microprocessors and
DSPs. Small program size and computational speed was very
convenient for miniaturization. The desired input and output
relationships were generally obtained with little compromise in
our case studies when we used FL. ANFIS was found more
convenient to obtain the desired relationship between the inputs
and the output.
We proposed the use of FL for development of an integrated
system that reduces the agility of the plane when any component
of the plane has problems. For small UAVs, an easy approach
was to evaluate the condition of all the considered components
and to adjust the gains of the PDs and PIDs that control the
autonomous flight system. We easily implemented the desired
input and output relationships using conventional FL and
ANFIS.
















6. ACKNOWLEDGEMENTS
This project was partially funded by the Air Force Research
Laboratory (GS-23F-8049H, F33601-03-F-0060). Authors
would like to thank Mr. Mark Derriso for his valuable
contributions to the work. In addition, authors appreciate the
help of Mr. Carlos Velez.
7. REFERENCES
[1] Wu, H., Sun, D., Nonmembers, Zhu, H., and Zhou, Z. An
autonomous flight control strategy study of a small-sized
unmanned aerial vehicle. IEICE Transactions of
Electronics. E88-10, (2005), 1-4.
[2] Menon, P. K., Iragavarapu, V. R. Blending homing
guidance law using fuzzy logic. AIAA Guidance,
Navigation and Control Conference, Boston, MA (1998).
[3] Livchitz, M., Abershitz, A., Soudak, U., and Kandel, A.
Development of an automated fuzzy-logic-based expert
system for unmanned landing. Fuzzy Sets and Systems, 93,
(1998), 145-159.
[4] Mengali, G. The use of fuzzy logic in adaptive flight
control systems. The Aeronautical Journal. 104, (2000),
31-37.
[5] Rajasekhar, V., and Sreenatha, A. G. Fuzzy Logic
Implementation of Proportional Navigation Guidance, Acta
Aeronautica. 46, (2000), 17-24.
[6] Shioiri, H., and Ueno, S. Three-dimensional collision
avoidance control law for aircraft using risk function and
fuzzy logic. Trans. Japan Soc. Aero. Space Sci, 154,
(2004), 253 -261.
[7] Yingjie, L. and Baoshu, W. Design of an aircraft auto-
landing fuzzy controller based on ANFIS. Journal of
System Simulation. 16, (2004), 2580-2583.
[8] Raisinghani, S. C., and Ghosh, A. K. Parameter estimation
of an aeroelastic aircraft using neural networks. Sadhana.
25, (2000), 181-191.